Biased Bayesian learning with an application to the risk-free rate puzzle